AI usage generates a fragile codebase

	AI usage generates a fragile codebase(evanxmerz.com)
	627 points by plun9 1 day ago \| 601 comments
	tl;dr: A former Meta engineer argues that heavy AI code generation produces fragile codebases marked by shallow tests, overly complex code, declining human understanding, and PR volume that outpaces meaningful review. The symptoms, he claims, include rising incident rates, longer recovery times, user-visible errors missed by automated tests, and degraded UX. His conclusion: quality software still requires a human touch, both for users and maintainers.
	HN Discussion: ↓Article is outdated as newer AI models have resolved the cited issues ↑Code quality issues only matter once they manifest as problems later ~Pre-AI codebases were also fragile, so this isn't unique to AI ↓Software is becoming a black box and that's acceptable as long as it works ↑Mandated high AI code generation percentages for ratings is disturbing